[PATCH] Avoid use of uninitialised spinlock in EEH.

David Woodhouse dwmw2 at infradead.org
Thu Nov 17 00:38:24 EST 2005


On Thu, 2005-09-22 at 14:46 +0200, Arnd Bergmann wrote:
> How about making that a firmware feature bit? That would avoid the
> extra global exported symbol and be slightly more efficient for
> the pSeries-only configuration.

But ppc64_firmware_features isn't exported yet either.

Also, it wouldn't be a lot more efficient, since we set
FW_FEATURE_PSERIES_ALWAYS to zero, so it would never have the effect of
optimising out the check anyway.

If we were to attempt to fix that, we'd have to handle other platforms;
we'd have to take into account FW_FEATURE_{PMAC,MAPLE,CELL}_ALWAYS when
setting FW_FEATURE_ALWAYS. And even if we did that, we'd still probably
want to set the FW_FEATURE_EEH bit _only_ if EEH-capable devices were
found, rather than having it in FW_FEATURE_PSERIES_ALWAYS. So we still
wouldn't optimise out the check.

Paulus, the reason your G5 doesn't die with CONFIG_EEH set is because
you haven't got spinlock debugging. Try turning that on too :)

-- 
dwmw2




More information about the Linuxppc64-dev mailing list